Comparing tumour staging and grading systems: a case study and a review of the issues, using thymoma as a model

An Integrated Tree-Based Classification Approach to Prognostic Grouping with Application to Localized Melanoma Patients

We propose an integrated tree-based approach for prognostic grouping of localized melanoma patients. This approach incorporates the survival tree model with the agglomerative hierarchical clustering to group terminal subgroups with similar prognoses together. The Brier score is used to evaluate the goodness of fit and the k-fold cross-validation test is used to evaluate the reproducibility of the scheme for prediction. The proposed approach is applied to an American Joint Committee on Cancer (AJCC) localized melanoma data set and compared with the current AJCC staging system. This approach performs more efficiently than the standard tree methods and has made improvement over the current AJCC melanoma staging system.

Free PSA isoforms and intact and cleaved forms of urokinase plasminogen activator receptor in serum improve selection of patients for prostate cancer biopsy

Clinicians currently use simple cut-points, such as serum prostate-specific antigen (PSA) >or=4 ng/ml, to decide whether to recommend further work-up for prostate cancer (PCa). As an alternative strategy, we evaluated multivariable models giving probabilities of a PCa diagnosis based on PSA and several circulating novel biomarkers. We measured total PSA, free PSA (fPSA), fPSA subfractions (single-chain fPSA-I and multichain fPSA-N), total human glandular kallikrein 2 (hK2) and full-length and cleaved forms of soluble urokinase plasminogen activator receptor (suPAR) in pretreatment serum from 355 men referred for prostate biopsy. Age and total PSA were combined in a "base" regression model to predict biopsy outcome. We then compared this base model to models supplemented by various combinations of circulating markers, using concordance index (AUC) to measure diagnostic discrimination. PCa prediction was significantly enhanced by models supplemented by measurements of suPAR fragments and fPSA isoforms. Addition of these markers improved bootstrap-corrected AUC from 0.611 for a cut-point and 0.706 for the base model to 0.754 for the full model (p=0.005). This improved diagnostic accuracy was also seen in subanalysis of patients with PSA 2-9.99 ng/ml and normal findings on DRE (0.652 vs. 0.715, p=0.039). In this setting, hK2 did not add diagnostic information. Measurements of individual forms of suPAR and PSA isoforms contributed significantly to discrimination of men with PCa from those with no evidence of malignancy.

Comparative assessment of the 1992 and 2002 pathologic T3 substages for the prediction of biochemical recurrence after radical prostatectomy

BACKGROUND

The objective of this study was to compare the ability of the 1992 American Joint Committee on Cancer (AJCC) TNM staging system for nonorgan-confined prostate carcinoma (PCa) (pathologic T3 [pT3a], pT3b,and pT3c) to predict biochemical recurrence (BCR) after radical prostatectomy with its revision from 1997/2002 (pT3a and pT3b).

METHODS

The authors analyzed prospectively collected data from 971 consecutive patients with pT3 tumors who underwent radical prostatectomy alone at a single institution. According to the 1992 AJCC substages, 494 patients had pT3a PCa (51%), 85 patients had pT3b PCa (9%), 302 patients had pT3c PCa (31%) and 91 patients had pT3 PCa (9%) with metastatic lymph node invasion (LNI). Kaplan-Meier and Cox proportional hazards regression analyses were employed to assess the BCR rate using the preoperative prostate-specific antigen (PSA) level, surgical margin (SM) status, pathologic Gleason score, and LNI. The predictive accuracy of this "base" model was compared with models that added either the AJCC 1992 or the AJCC 1997/2002 staging definitions.

RESULTS

BCR was observed in 345 patients (36%). The actuarial 5-year recurrence-free probability for patients with AJCC 1992 pT3a, pT3b, and pT3c PCa was 69%, 42%, and 33%, respectively, and differed significantly between men with pT3a tumors and men with pT3b tumors (log-rank test; P < 0.0005). In Cox regression analyses, the 1992 substages were associated significantly with BCR after controlling for PSA, SM status, Gleason grade, and LNI (P < 0.0005). The predictive accuracy of the base model (bootstrap-corrected concordance index, 0.739) was improved after addition of the 1992 TNM staging criteria (concordance index, 0.747). However, predictive accuracy was similar for the model that included the 1997/2002 substages (concordance index, 0.746).

CONCLUSIONS

Substaging of pT3 tumors according to the less complex 1997/2002 pT3 classification improved the predictive accuracy of the BCR rate virtually to the same extent as the more detailed 1992 classification. Therefore, the current data favor the use of the 1997/2002 pT3 substaging system.

Estimates of stage-specific survival are altered by changes in the 2002 American Joint Committee on Cancer staging system for melanoma

BACKGROUND

The objectives of the current study were to examine how the estimated stage-specific survival is altered in the 2002 American Joint Committee on Cancer (AJCC) melanoma staging system compared with the 1997 AJCC staging system and to contrast the predictive accuracy of the 2 staging systems.

METHODS

There were 5847 consecutive melanoma patients who presented to Memorial Sloan-Kettering Cancer Center from 1996 to 2004 and who were entered prospectively into a data base. These patients were staged according to both the 1997 and 2002 AJCC staging criteria. Overall survival estimates were determined using the Kaplan-Meier method. The overall predictive accuracy of the two staging systems was compared using concordance estimation.

RESULTS

In total, 1035 patients were shifted to a lower stage in the 2002 staging system, whereas only 15 patients were upstaged. The number of patients with Stage I melanoma increased by 697 under the 2002 system (n = 2166 patients) compared with the 1997 system (n = 1463 patients). Because of the changes in 2002, the estimated 5-year overall survival for patients with Stage II melanoma decreased considerably, from 79% (1997) to 64% (2002). With the initiation of subgroups in 2002, it became apparent that patients with Stage III melanoma were very heterogeneous in terms of their survival probabilities (5-yr overall survival ranged from 70% in patients with Stage IIIA disease to 24% in patients with Stage IIIC disease). Furthermore, in the 2002 system, there was substantial prognostic overlap between Stage II and Stage III. Despite the increased complexity of the 2002 system, the 2 staging systems had similar concordance estimates: 58% for the 1997 staging system compared with 58% (ignoring the subgroups) and 59% (with subgroups) for the 2002 system.

CONCLUSIONS

Estimates of stage-specific survival were altered substantially by the changes made in the 2002 AJCC staging system for melanoma, particularly for Stage II. Stage subgroups that were added in the 2002 system resulted in a large diversity of risk within Stage III. This must be taken into account to stratify patients properly for clinical trials. The increased complexity of the 2002 system did not improve its predictive ability over the simpler 1997 system, highlighting the importance of developing individualized risk-prediction models.

Tumor size improves the accuracy of TNM predictions in patients with renal cancer

OBJECTIVES

Current staging for renal cancer (RC) does not directly rely on tumor size. We examined the increment in accuracy related to inclusion of pathologically determined tumor size in prediction of nodal metastases (N+), distant metastases (M+), and cancer-specific survival (CSS).

METHODS

Partial or radical nephrectomy was performed in 2245 patients with clear cell histology. Pathologic stages were T1a in 566, T1b in 490, T2 in 303, T3 in 831, and T4 in 55 patients. Tumor size was 0.5-25 cm (mean, 6.8). Multivariate models relied on 1997 and 2002 TNM variables and addressed N+, M+ disease, and CCS. Their accuracy was compared according to either the presence or absence of tumor size.

RESULTS

In all univariate and multivariate models, tumor size was a statistically significant predictor of all outcomes (p< or =0.001). In all multivariate models, tumor size added between 3.7% and 0.8% to predictive accuracy of either 1997 or 2002 TNM categories.

CONCLUSIONS

Tumor size represents a highly significant, multivariate, and informative predictor of RC outcomes and may warrant inclusion in future TNM revisions.

Which classification for ethmoid malignant tumors involving the anterior skull base?

BACKGROUND

The purpose of this study was to compare three systems of classification for malignant ethmoidal tumors in patients undergoing anterior craniofacial resection.

METHODS

A radiologic locoregional evaluation of 241 patients with malignant ethmoid tumors was performed before patients underwent an anterior craniofacial resection. Disease in each case was staged according to the American Joint Committee on Cancer-Union Internationale Contre le Cancer (AJCC-UICC) 1997 classification, the AJCC-UICC 2002 classification, and the Istituto Nazionale Tumori (INT) classification. Kaplan-Meier curves and Cox models were used to investigate the prognostic value of each classification system on disease-free survival (DFS) and overall survival (OS). The classifications were compared in terms of prognostic discrimination capability, measured by use of an index of agreement between each classification and DFS or OS time.

RESULTS

All three classification systems yielded statistically significant results in the Cox analysis, both for DFS and OS. In the AJCC-UICC 2002 system, minor differences were observed between T1 and T3 tumors. The INT classification showed a progressive worsening of the prognosis with increasing stage. The index of prognostic discrimination favored the INT classification over both the 1997 and 2002 AJCC-UICC classifications.

CONCLUSIONS

Both the 1997 and 2002 AJCC-UICC classifications seemed to have limited prognostic value. By contrast, the INT classification satisfied one of the main goals of tumor staging, demonstrating the progressive worsening of prognosis with different tumor classes.

A nomogram that predicts the presence of sentinel node metastasis in melanoma with better discrimination than the American Joint Committee on Cancer staging system

BACKGROUND

The threshold and indications for sentinel lymph node (SLN) biopsy in patients with melanoma remain somewhat arbitrary. Many variables associated with SLN positivity have previously been identified, including a significant association between the American Joint Committee on Cancer (AJCC) staging system and SLN status. We developed a user-friendly nomogram that takes several characteristics into account simultaneously to more accurately predict the presence of SLN metastasis for an individual patient.

METHODS

A total of 979 patients who underwent successful SLN biopsy for cutaneous melanoma at a single institution between February 1991 and November 2003 were included in the analysis. Predictors were used to develop a nomogram, based on logistic regression analysis, to predict the probability of SLN positivity. A large multi-institutional trial with 3108 patients was used to validate the predictive accuracy of the nomogram compared with the AJCC staging system.

RESULTS

The nomogram was developed and found to be accurate and discriminating. The concordance index of the nomogram, a measure of predictive ability, was .694 when evaluated with the validation dataset. In contrast, the concordance index of the AJCC staging system was lower (.663; P < .001).

CONCLUSIONS

Using commonly available clinicopathologic information, we developed a nomogram to accurately predict the probability of a positive SLN in patients with melanoma. This tool takes several characteristics into account simultaneously. This model should enable improved patient counseling and treatment selection.

The International Bladder Cancer Bank: proposal for a new study concept

At present, results of marker studies are often inconsistent and sometimes contradictory. Recognized problems include multiple different methods of performing the assays, different subsets of patients and different endpoints, leading to incompatible datasets. Although there has been discussion of establishing general methodological principles and guidelines (analogous to those for clinical trials) for design, conduct, analysis, and reporting of marker studies, these have not been widely implemented. There are no well-recognized prototypes or examples that the urologic researcher can use to model future marker studies. We will discuss our plans to establish a multi-institutional bladder cancer data base and virtual tumor bank as a resource for participating institutions to evaluate the biological and prognostic significance of potential markers for bladder cancer. Samples will be identified and stored at each participating institution and will be available for analysis. A standard, minimal set of patient and pathologic information will be collected. The use of common software, as part of this proposal will facilitate the data transfer of updated patient information to a central database. All contributing centers will have access to summarized information, also to simplify the process of finding collaborating partners. Prospectively collected, consistent datasets with available long-term follow-up, should provide information sooner than with a conventional prospective study. Furthermore, the quality of these data and samples may be superior to that of retrospectively collected data and samples. The proposed International Bladder Cancer Bank of specimens and data will be an effective tool during all phases of marker development.

Confusion by frequent changes in staging of exocrine pancreatic carcinoma

OBJECTIVES

The TNM/pTNM classification of anatomic extent before treatment is the strongest predictor of outcome in exocrine pancreatic carcinoma. Frequent changes in staging, published by the UICC in 1987, 1997, and 2002, lead to considerable problems.

METHODS

The data on 272 patients with resection of a pancreatic ductal adenocarcinoma between 1978 and 1997 were analyzed.

RESULTS

Two hundred sixty-five tumors were assigned to a higher pT category in 1997. Of them, 70 were reassigned to a lower pT category in 2002. No patient fulfilled the criteria of pT4 in 2002. Eighty-seven tumors were assigned to a higher pathologic stage in 1997. In 2002, 151 tumors were assigned to a lower pathologic stage. No patient was assigned to pathologic stage III. The staging systems of 1987 and 1997 are able to identify subgroups of patients with superior prognosis. The staging system of 2002 includes the same 12 patients in stage I as the classification of 1997. However, stage II contains an inhomogeneous group of 193 patients with poor prognosis.

CONCLUSIONS

Changes in the TNM classification require a conversion of the data. Analysis and comparison of published results are very difficult and sometimes impossible if classification systems change too often. The present classification is well qualified for treatment choice and gives good information on prognosis after resection. It should be unchanged for at least 10 years.

Validation of the Kattan preoperative nomogram for prostate cancer recurrence using a community based cohort: results from cancer of the prostate strategic urological research endeavor (capsure)

PURPOSE

The Kattan preoperative nomogram combines preoperative prostate specific antigen (PSA), biopsy Gleason grade and clinical stage to estimate disease recurrence after radical prostatectomy. Several studies using patient data from academic centers have validated the nomogram. We assessed the performance of the Kattan nomogram using the Cancer of the Prostate Strategic Urological Research Endeavor database, a national, largely community based observational disease registry.

MATERIALS AND METHODS

From the Cancer of the Prostate Strategic Urological Research Endeavor database we identified 1701 men with clinically localized prostate cancer undergoing radical prostatectomy with sufficient pretreatment information and PSA followup after surgery. Disease recurrence was defined as 2 consecutive PSA values 0.2 ng/ml or greater, or a second cancer treatment more than 6 months after prostatectomy. A concordance index was used to evaluate the performance of the nomogram compared to observed 5-year recurrence-free survival (Kaplan-Meier). Kattan nomogram scores were calculated for each patient and stratified into 6 groups for analysis.

RESULTS

In our cohort of 1701 men 413 (24%) had evidence of disease recurrence. Median followup in these patients was 2.3 years. Kattan nomogram scores were 17% to 99% (mean 79%). The overall concordance index was 0.68. Varying the definition of recurrent disease and excluding patients with imputed data did not substantially alter nomogram performance (concordance index 0.65 to 0.70). The Kattan nomogram tended to overestimate 5-year freedom from recurrence in patients with scores of 65% and higher.

CONCLUSIONS

We noted the reasonable performance of the Kattan nomogram for predicting cancer outcomes after radical prostatectomy using a community based population. Although concordance is lower than in previous validation studies and the nomogram overestimates recurrence-free survival in patients at lower risk, the model is fairly robust and it provides important information when counseling patients regarding treatment options in the community setting. Further refinements in pretreatment estimation of disease-free survival and ultimately overall survival are needed.

Assessment of the Enhancement in Predictive Accuracy Provided by Systematic Biopsy in Predicting Outcome for Clinically Localized Prostate Cancer

PURPOSE

Current localized prostate cancer treatment outcome nomograms rely on prostate specific antigen (PSA), tumor stage and grade. We investigated whether the addition of prostate biopsy features may enhance the accuracy of a nomogram predicting recurrence after radical prostatectomy (RP).

MATERIALS AND METHODS

Clinical data from 1,152 patients who underwent RP were used and included PSA, clinical stage, biopsy Gleason grade and systematic biopsy information that quantified the amount of cancer and high grade cancer. Predictive accuracy for freedom from recurrence after RP was assessed with and without tumor quantification in the biopsy by the area under the receiver operating characteristics curve (AUC).

RESULTS

Percentage and number of cores with cancer, and percentage and number of cores with high grade cancer were predictors of outcome when added to models that included PSA, Gleason grade and clinical stage (all p <0.0001). Nomogram accuracy with 3 traditional variables (AUC 0.790) was minimally enhanced with the addition of percentage or number of positive cores (AUC 0.804 and 0.800, respectively), or percentage or number of cores with high grade cancer (AUC 0.802 and 0.800, respectively). Maximum predictive accuracy of 0.811 was achieved after supplementing the traditional 3-variable nomogram with various combinations of additional pathological predictors.

CONCLUSIONS

The information provided by systematic biopsies substantially improves the ability to predict outcome following RP. However, some incremental predictive accuracy was achieved by adding systematic biopsy features.

Use of nomograms to predict the risk of disease recurrence after definitive local therapy for prostate cancer

The generally indolent nature of prostate cancer, as well as the impact that treatment can have on quality of life (QOL) and cancer control, makes the decision analysis difficult for patients facing the task of selecting a treatment for clinically localized disease. Instruments to aid patients and their physicians in this decision analysis are needed. Nomograms are instruments that predict outcomes using specific clinical parameters. Nomograms use algorithms that incorporate several variables to calculate the predicted probability that a patient will achieve a particular clinical end point. Nomograms tend to outperform both clinical experts and predictive models using methods of risk grouping. We briefly outline the uses and limitations of nomograms, principles of nomogram construction, and the available models for predicting the progression-free probability after local definitive therapy with radical prostatectomy, external-beam radiotherapy, or brachytherapy. There is a need for additional nomograms that predict outcomes after salvage therapy, as well other clinical end points, including QOL-adjusted survival.

Comparison of Cox regression with other methods for determining prediction models and nomograms

PURPOSE

There is controversy as to whether artificial neural networks and other machine learning methods provide predictions that are more accurate than those provided by traditional statistical models when applied to censored data.

MATERIALS AND METHODS

Several machine learning prediction methods are compared with Cox proportional hazards regression using 3 large urological datasets. As a measure of predictive ability, discrimination that is similar to an area under the receiver operating characteristic curve is computed for each.

RESULTS

In all 3 datasets Cox regression provided comparable or superior predictions compared with neural networks and other machine learning techniques. In general, this finding is consistent with the literature.

CONCLUSIONS

Although theoretically attractive, artificial neural networks and other machine learning techniques do not often provide an improvement in predictive accuracy over Cox regression.

Prognostic factors for survival in patients with hormone-refractory prostate cancer (HRPC) after initial androgen deprivation therapy (ADT)

Androgen deprivation therapy (ADT) is a standard mode of therapy for patients with metastatic prostate cancer. Controversy exists, however, as to the optimal timing of initiation of ADT, as well as whether this form of therapy imparts a survival benefit to patients with advanced disease. Side effects of ADT are not minimal and can seriously compromise a patient's quality of life. Additionally, ADT eventually results in hormone-refractory prostate cancer (HRPC). Despite new chemotherapeutic regimens and hormonal agents, overall survival in these patients remains universally low. Nonetheless, it is valuable to gauge a patient's prognosis to assist in decision making when considering treatment options. Contemporary series analyzing patients with HRPC have identified several factors prognostic of survival outcomes, such as lactate dehydrogenase (LDH), alkaline phosphatase (ALK), hemoglobin (Hgb), and serum prostate specific antigen (PSA) level. Nomograms have been developed that utilize these pretreatment clinical variables to predict clinical outcomes, including 1-year, 2-year, and median survival times in patients with HRPC. These instruments are capable of more accurately predicting survival outcomes than traditional tables of multivariate results or simple analysis of prognostic factors. We believe these nomograms will become indispensable tools for patient counseling and clinical trial design in patients with HRPC.

Nomograms are superior to staging and risk grouping systems for identifying high-risk patients: preoperative application in prostate cancer

PURPOSE OF REVIEW

We outline a generic approach to using a nomogram to predict a continuous probability of failure in high-risk patients (rather than putting patients into groups), in order to identify patients whose risk exceeds a cutoff point. We discuss the goals of any staging system, what markers should be included, and models of markers.

RECENT FINDINGS

Selection of high-risk patients for any cancer has traditionally been accomplished by the creation of risk groups, or perhaps clinical stages. Ideally, high-risk patients should be identified as accurately as possible, because of the treatment and psychological implications for the patient. We argue that a continuous multivariable prediction model, such as a nomogram, is the most appropriate and accurate way to select high-risk patients. This type of model predicts outcome more accurately than risk grouping or staging systems. As an example, we use our preoperative prostatic specific antigen recurrence nomogram to identify patients at high risk of biochemical failure, who are in need of an effective neoadjuvant therapy.

SUMMARY

It will follow from our discussion that identification of high-risk patients should follow four simple steps. First, select the endpoint of interest for the trial or the patient. Second, select the method that predicts the endpoint as accurately as possible. Third, determine the cutoff of predicted probability beyond which it makes sense to give the patient experimental therapy. Fourth, offer the novel therapy to the patient whose prediction of the endpoint, using the most accurate prediction method, exceeds the threshold.

Prediction of Progression: Nomograms of Clinical Utility

It is difficult to determine the pathologic stage of a clinically localized prostate cancer by physical examination or imaging studies. Consequently, clinicians rely on predictive models that estimate the probability of lymph node metastases and other pathologic features from clinical factors such as the clinical T stage, the grade in the biopsy specimen, and the serum prostate-specific antigen level. These models do not, however, directly predict prognosis. In developing a tool for predicting the probability that prostate cancer might recur after treatment, we took a novel approach that focused on the risk for the individual patient. In particular, we chose to develop a tool that calculates a continuous probability of recurrence rather than placing the patient in a risk group. This represents a fundamental departure from the classical goal of staging; a departure we argue is long overdue. Clinically localized prostate cancer patients deserve the most accurate and tailored predictions available, which current staging systems do not provide. Such an individualized approach should add value in medical decision making whenever an accurate prediction of the outcome may guide treatment selection.

International validation of a preoperative nomogram for prostate cancer recurrence after radical prostatectomy

PURPOSE

We evaluated the predictive accuracy of a recently published preoperative nomogram for prostate cancer that predicts 5-year freedom from recurrence. We applied this nomogram to patients from seven different institutions spanning three continents.

METHODS

Clinical data of 6,754 patients were supplied for validation, and 6,232 complete records were used. Nomogram-predicted probabilities of 60-month freedom from recurrence were compared with actual follow-up in two ways. First, areas under the receiver operating characteristic curves (AUCs) were determined for the entire data set according to several variables, including the institution where treatment was delivered. Second, nomogram classification-based risk quadrants were compared with actual Kaplan-Meier plots.

RESULTS

The AUC for all institutions combined was 0.75, with individual institution AUCs ranging from 0.67 to 0.83. Nomogram predictions for each risk quadrant were similar to actual freedom from recurrence rates: predicted probabilities of 87% (low-risk group), 64% (intermediate-low-risk group), 39% (intermediate-high-risk group), and 14% (high-risk group) corresponded to actual rates of 86%, 64%, 42%, and 17%, respectively. The use of neoadjuvant therapy, variation in the prostate-specific antigen recurrence definitions between institutions, and minor differences in the way the Gleason grade was reported did not substantially affect the predictive accuracy of the nomogram.

CONCLUSION

The nomogram is accurate when applied at international treatment institutions with similar patient selection and management strategies. Despite the potential for heterogeneity in patient selection and management, most predictions demonstrated high concordance with actual observations. Our results demonstrate that accurate predictions may be expected across different patient populations.

Statistical characteristics of area under the receiver operating characteristic curve for a simple prognostic model using traditional and bootstrapped approaches

Prognostic models are increasingly common in the biomedical literature. These models are frequently evaluated with respect to their ability to discriminate between those with and without an outcome. The area under the receiver-operating curve (AROC) is often used to assess discrimination. In this study, we introduce a bootstrap method, and, using Monte Carlo simulation, we compare three different bootstrap approaches with four commonly used methods in their ability to accurately estimate 95% confidence intervals (CIs) around the AROC for a simple prognostic model. We also evaluated the power of a bootstrap method and the commonly used trapezoid rule to compare different prognostic models. We show that several good methods exist for calculating 95% CIs of AROC, but the maximum likelihood estimation method should not be used with small sample sizes. We further show that for our simple prognostic model a bootstrap z-statistic approach is preferred over the trapezoidal method when comparing the AROCs of two related models.

Analysis of clinical stage T2 prostate cancer: do current subclassifications represent an improvement?

PURPOSE

The purpose of this study was to determine whether the extent of palpable cancer within the prostate predicts outcome after radical prostatectomy.

PATIENTS AND METHODS

We combined prospectively collected data on 1,755 consecutive clinical stage T2 patients treated with radical prostatectomy alone at four institutions. According to the 1992 American Joint Committee on Cancer tumor-node-metastasis system, 645 (37%) were T2a, 758 (43%) were T2b, and 352 (20%) were T2c. Kaplan-Meier and proportional hazards regression analyses were performed on the 1992 and 1997 T2 subclassifications. After controlling for the effects of prostate-specific antigen (PSA) and biopsy Gleason sum, the two staging systems were compared for their ability to predict recurrence-free survival (RFS). Adjusted RFS curves were constructed using the corrected group prognosis method.

RESULTS

Follow-up ranged from 1 to 166 months (median, 26 months). Cancer recurred in 417 (24%) of the T2 patients. The 1992 (P =.005) but not the 1997 (P =.100) T2 subclassification predicted outcome after controlling for PSA and Gleason sum. After covariate adjustment, RFS was 7% higher at 5 years in the 1992 T2a subcategory relative to the T2b subcategory.

CONCLUSION

The 1992 American Joint Committee on Cancer system is superior to the 1997 system, and the former adds prognostic information to a model containing pretreatment PSA and Gleason sum. These results suggest that 1992 T2 subclassification derived from palpable findings improves prognostication over the 1997 subclassification.

Validation study of the accuracy of a postoperative nomogram for recurrence after radical prostatectomy for localized prostate cancer

PURPOSE

A postoperative nomogram for prostate cancer was developed at Baylor College of Medicine. This nomogram uses readily available clinical and pathologic variables to predict 7-year freedom from recurrence after radical prostatectomy. We evaluated the predictive accuracy of the nomogram when applied to patients of four international institutions.

PATIENTS AND METHODS

Clinical and pathologic data of 2,908 patients were supplied for validation, and 2,465 complete records were used. Nomogram-predicted probabilities of 7-year freedom from recurrence were compared with actual follow-up in two ways. First, the area under the receiver operating characteristic curve (AUC) was calculated for all patients and stratified by the time period of surgery. Second, calibration of the nomogram was achieved by comparing the predicted freedom from recurrence with that of an ideal nomogram. For patients in whom the pathologic report does not distinguish between focal and established extracapsular extension (an input variable of the nomogram), two separate calculations were performed assuming one or the other.

RESULTS

The overall AUC was 0.80 when applied to the validation data set, with individual institution AUCs ranging from 0.77 to 0.82. The predictive accuracy of the nomogram was apparently higher in patients who were operated on between 1997 and 2000 (AUC, 0.83) compared with those treated between 1987 and 1996 (AUC, 0.78). Nomogram predictions of 7-year freedom from recurrence were within 10% of an ideal nomogram.

CONCLUSION

The postoperative Baylor nomogram was accurate when applied at international treatment institutions. Our results suggest that accurate predictions may be expected when using this nomogram across different patient populations.